From: Han-Wen N. <ha...@gm...> - 2012-05-28 03:45:50
|
On Mon, May 28, 2012 at 12:38 AM, Liu Yuan <nam...@gm...> wrote: > Our project implements a fuse client to access a distributed storage. > We can change FUSE_MAX_PAGES_PER_REQ to bigger value to get the expected > read size(say, 4M) in one go, but the write size is still limited to > 128K. The request size matters because our distributed storage is an > object storage with the object size much bigger than 128K. > We don't want the client request to be split, because split requests > will be considered multiple requests to the distributed storage, for > example, with object size 4M, a 4M request will be split into 32 storage > requests, the overhead is considerably large. Why don't you sync with the backend when the client closes the file, ie. when the fuse daemon handles the release callback? You then have the option of using as large chunks as you want. -- Han-Wen Nienhuys - ha...@xs... - http://www.xs4all.nl/~hanwen |